Tekst finds the work your software never sees

An order exists before it becomes an order. It sits in an email, perhaps with a PDF attached, while somebody checks a product code, asks a colleague about delivery and waits for a customer to clarify a quantity. Then it enters SAP. A dashboard starts its clock. Several hours of work have already disappeared from the picture.

This is the opening Tekst has chosen. The Belgian software company treats the correspondence surrounding a transaction as operational evidence. Read the messages, connect them to records in other systems, and a different account of the business emerges: the actual routes, the waiting, the exceptions, the people holding everything together.

The tyranny of the routing rule

Consider Securex, the HR and payroll business. In Tekst’s account of the deployment, Securex faced 2.3 million emails a year and more than 1,700 hard-coded routing rules. Only a third of messages reached the correct teams automatically. The rest needed dispatching by hand. An elaborate system had become a very busy invitation to do more work.

“Email handling was consuming too much of our resources,” said Christophe Lapeau, Securex’s Salesforce product owner. Tekst added contextual classification: language, urgency, complaint status. During testing, the case study reports routing accuracy increasing from 50% to 90% within weeks. That accuracy comparison is a separate measure from the earlier proportion routed automatically.

The amusing consequence was that Securex identified more complaints. Better detection made an unpleasant number grow. Yet the company welcomed the information because it could address problems earlier. Sometimes the first benefit of better software is an uglier, more useful view of reality.

A map before a motor

Founded in 2022 by CEO Wouter Janssen and CTO Tiebe Parmentier, Tekst initially attracted attention for turning unstructured text into usable business data. In May 2024, it secured €700,000 from Entourage Capital for development and European commercial expansion. Medical supplier Becton Dickinson and dairy company Milcobel were already named customers.

Its present proposition has two linked parts. Process Intelligence reconstructs how requests move across people and systems. Agentic Process Automation acts on that understanding, routing a message, updating a case or processing an order. Custom-trained models learn business terminology and context from historical and live interactions.

The distinction matters in a crowded market. Celonis offers process mining; UiPath supplies robotic process automation; Esker tackles document and transaction workflows. SAP and Salesforce have their own automation capabilities. Tekst’s chosen ground is the unstructured communication connecting those systems. Its argument is that a transaction log can reveal an event while missing the conversation that explains it.

Universal Tracing, labelled patent pending, links records belonging to the same process across different identifier schemes. SidePanel gives employees the source document beside extracted data and proposed actions. They can approve, correct or reject. That is a practical acknowledgement that handing work to software can happen one decision at a time.

“You can’t automate what you can’t see.”

Tekst’s operating proposition

Twenty seconds, repeated two million times

Daikin Europe offers a larger test of the idea. Its customer-service and order teams manually read, categorized and routed tickets across regions. Processes differed, and existing process-mining tools provided limited visibility. Tekst first analyzed live tickets and events, with business input, to establish which improvements deserved priority.

Classification and assignment followed. The case study reports more than two million tickets auto-classified, over twenty seconds saved per ticket and 1,800 working days of manual handling removed. These are Tekst’s reported deployment results. The apparently tiny saving is the point: repetitive enterprise work makes seconds consequential.

The same exercise identified order intake and quote work as subsequent opportunities. Those belong to the roadmap, rather than the completed savings. That distinction is useful for anyone buying AI: an identified pool of work is potential; a changed operating process is evidence.

At Dossche Mills, the less glamorous materials are flour orders, multilingual mailboxes and complaints. Tekst connects Outlook, SAP and SharePoint to categorize requests, route them and support order handling. The company reports maintaining order volume with a reduced workforce. It is a recognisable enterprise problem: experienced people leave, but the inbox does not retire.

Pay for the work, check the arithmetic

Tekst sells enterprise software through tailored quotes. Its current pricing separates analyzed event volume from automation credits. Moving a message consumes fewer credits than creating a case; creating a case consumes fewer than booking a sales order. Credits are purchased upfront under annual contracts, test runs are free, and usage is visible in the product.

That makes the buying exercise fairly concrete. Estimate the incoming volume, test the extraction and routing, count the actions required, then compare the charge with handling time and errors removed. Include the cost of reviewing exceptions. A cheap prediction that creates an expensive correction is a poor bargain.

In May 2026, Tekst announced an €11.5 million Series A led by Elephant, with Entourage participating. The disclosed plans covered product development, international expansion and doubling a team of 35 by year-end. Public vacancies span engineering and commercial roles. Growth capital buys the opportunity to execute those plans; it does not prove them completed.

The useful thing to steal

The transferable lesson is simple enough to try without buying anything. Take one recurring request. Follow it from the first message to completion. Count the handoffs and corrections. Establish what a successful automated action must do, and keep a person available for consequential exceptions.

As an operating principle, this fits repeated, high-volume work with accessible communications and connected systems. A sparse workflow, inaccessible history or a decision needing unusual human judgement makes the economics harder. Tekst’s interesting wager is that understanding the ordinary mess comes first. The inbox, for once, gets to explain itself.